Aim To examine how gender, self-efficacy, attachment styles, and social influence AI dependence among college

students.

Background Artificial Intelligence (AI) has become increasingly important in daily life, and researchers have

identified psychological factors as significant determinants of AI dependence.

Methods A total of 154 college students aged between 18 and 28 were selected. Data was collected through a

questionnaire, and participants’ AI Usage, Self-Efficacy, Attachment style, and Dependency were assessed

through scales based on the Technology Acceptance Model (extended TAM), New General Self-Efficacy Scale

AISES subscales. DAI showed significant positive correlations with AISE-AS.

Conclusion

Males showed higher AI-dependency, while attachment styles were not significantly related. Human-like

interaction and perceived trust in AI predicted AI Dependence.

Keywords: Artificial Intelligence (AI), Psychological factor, Gender differences, College Students, AI-

dependency

The rapid advancement in technology has led to an increase in the usage of AI in various facets of life, including

education (Yilmaz & Karaoğlan Yilmaz, 2023).AI has given significant competitive advantages to those who

have knowledge about its utilization and could be used to create new innovative products and services

significantly impacts an individual’s approach to the goal, measurement, and sources of self-efficacy have been

extensively studied (Heslin & Klehe, 2008), and standardized instruments such as the New General Self-Efficacy

Scale (NGSES) (Chen & Gully, 2001) were developed.

It was found that in previous studies, familiarity and skill in using computer programs contribute to the

development of general computer self-efficacy (Iverson, Brooks, Ashton, Johnson, & Gualtieri, 2009). The

emergence of AI, a new Self efficacy has also emerged: AI self-efficacy, which refers to an individual's

confidence in their ability to effectively interact and utilize AI systems. The development of the validation of

scale AISES (Wang and Chuang,2023) is important for assessing AI efficacy. The AI-Self efficacy is dependent

on AI attitude, interest, and anthropic factors of AI and even on social influence (Hong,2021).

Beyond Self-efficacy, other psychological dimensions like Attachment style also cause an impact on AI (Wu et

The extended TAM construct provides an efficient framework for predicting user acceptance of new

technologies, emphasizing the significance of factors like perceived usefulness and perceived ease of use

(Baitekov, 2023). TAM bridges the users' psychological intentions with their practical evaluations of a

technology’s usefulness and ease of use, while remaining adaptable to different contexts. Social influence is an

important factor in technology adoption. (Dahri and Yahaya,2024).

Dependence on AI refers to the over-reliance on an AI system for performing tasks. Dependence on AI becomes

problematic when users start to prefer AI for everyday tasks, which leads to reduced autonomy and self-

confidence. (Zhang and Zhao, 2024) It may also result in a decline in critical thinking and creativity. (Gerlich,

2025) The scale for dependence on artificial intelligence (DIA) is important to assess AI dependence. (Morales-

García & Sairitupa-Sanchez, 2024)

and to reduce potential negative outcomes. Therefore, this study examines the psychosocial determinants to fill

the research gap, providing valuable insights into human AI interaction.

In this study, a cross-sectional quantitative research design was employed. The research was conducted on

university students with the help of a self-reported questionnaire. The study employed an offline method of data

collection through a hard copy of the self-reported questionnaire administered to the participants.

The participants from various disciplines are considered for the study, mainly undergraduate and postgraduate

students. Participants were informed about the study's purpose, and they were assured of the confidentiality of

their identity and data. Participants were given a questionnaire, and the process of filling out the questionnaire

Instruments: In the present study, standardized and validated self-report measures are employed to measure

psychological variables like self-efficacy, AI acceptance, dependency on AI, and attachment patterns. The

following tools were used:

1. New General Self-Efficacy Scale (NGSES) General Self-efficacy was measured with the help of the New

General Self-Efficacy Scale. The scale has 8 items that assess individual competence in various

situations. The items were scored on a 5-point Likert scale ranging from 1(strongly disagree) to

5(strongly agree), with higher ratings indicating greater self-efficacy. (Chen & Gully, 2001)

2. AI Self-Efficacy Scale (AISES) AI self-efficacy was measured using the AISES, which measures

individuals' confidence in collaborating with and applying AI tools. This scale has four sub-scales which

are- Assistance (AISE-AS), Anthropomorphic interaction (AISE-AI), Comfort with AI(AISE-CF), and

scores indicating greater dependency. (Morales-García & Sairitupa-Sanchez, 2024)

4. Technology Acceptance Model (extended TAM) for AI Acceptance of AI was developed from an

extended version of the Technology Acceptance Model (extended TAM) (Lin et al., 2023), which was

adapted for the measurement of ChatGPT acceptance in educational environments. It has subscales for

the measurement of Personal Competence (extended TAM-PC), Social Influence (extended TAM-SI),

Perceived AI Trust (extended TAM-PAI), Perceived Usefulness of AI (extended TAM-AIU), Perceived

AI Enjoyment (extended TAM-AIE), Perceived Usefulness (extended TAM-PAII), Attitude towards

ChatGPT (extended TAM-Ach), Metacognitive Self-Regulated Learning (extended TAM-SRL) (Dahri

and Yahaya,2004).

5. Experiences in Close Relationships-Revised (ECR-R) Attachment styles were measured using the ECR-

current study.

The collected data were analyzed with the help of Statistical Package for the Social Sciences (SPSS) v. 31.

Firstly, the data were screened for missing values and outliers. Questionnaires with incomplete responses were

also excluded from the final analysis.

Descriptive analysis, such as means, standard deviation, and frequency, was calculated to summarize participant

demographics and the central tendencies of variables, including AI dependence, general self-efficacy, AI self-

efficacy, attachment styles, and various extended TAM constructs.

Inferential statistical analyses were conducted to address the research objectives:

The final sample consisted of 134 college students aged between 18 and 28 years (M = 21.40, SD = 2.63), with

89 males (65.93%) and 46 females (34.07%). Descriptive statistics for variables and their subscales: General

Self-Efficacy (NGSES), AI Self-Efficacy (AISES), Dependence on AI (DAI), Attachment Styles (ECR-R), and

the extended TAM construct are performed.

An independent samples t-test was conducted to determine the impact of gender differences in dependence on

AI (Table.1). The results indicated that males (M = 14.06, SD = 3.77, n = 88) had a significantly higher mean

score on the Dependence on Artificial Intelligence (DAI) scale compared to females (M = 12.33, SD = 4.36, n =

46).

The difference in means was statistically significant (t =2.28, p < .05), with a mean difference of 1.73 (95% CI

Table: 3.1 Significant Positive Correlations of Dependence on AI (DAI) with AI Self-Efficacy and extended

TAM Constructs (N=134)

DAI also exhibited significant positive correlations with multiple extended TAM constructs, including extended

extended TAM-PAI = Perceived AI Trust from the AI-Technology Acceptance Model.

**p < .01

Among all the predictors:

Anthropomorphic Interaction (AISE-AI) was a significant positive predictor of AI dependence,

dependent on it.

Perceived AI Trust (extended TAM-PAI) was a significant negative predictor,

dependence.

Other predictors, including general self-efficacy, emotional assistance, comfort with AI, technological skill,

perceived enjoyment, attitude, intention, and attachment dimensions, do not significantly contribute to the model

(p > .05).

dependence on AI, evidenced by their higher average scores relative to females; this distinction has a small to

medium effect size. Research indicates that women tend to have a less favorable attitude toward AI, engage with

AI technologies less frequently, possess lower perceived knowledge of AI, and report higher levels of AI-related

anxiety (Otis et al., 2025; Russo, 2025). This could create a psychological landscape where women are less likely

to form a strong reliance or dependence on AI technologies. This finding highlights the need for more gender

specific studies of AI interaction. Though no significant gender difference was observed for NGSES and AI-Self

Efficacy, it was found in earlier research that women have lower General self-efficacy and AI-Self Efficacy

compared to males (Asio and Sardina, 2025; Otis et al., 2025).

There was no significant correlation found between General Self-Efficacy (NGSES) and Dependence on

AI(DAI) in the current study, which does not align with the previous research, as a significant relation was found

social influence, trust, perceived usefulness, enjoyment, positive attitudes, self-regulation, and intention to use

AI. This establishes Bandura's self-efficacy theory, suggesting that confidence in one's ability to effectively

interact with and utilize AI systems directly contributes to their integration and reliance (Bandura

1977,1986,1997). Research indicates that an increase in the usage of AI, such as generative AI, can increase

efficiency and confidence, enhancing self-efficacy (Liang et al., 2023b; Yilmaz and Yilmaz, 2023), and the use

of AI may lead to dependence on AI, which weakens students' ability to solve problems independently.

(Octaberlina et al., 2024; Zhang et al., 2024)

The various dimensions of AI self-efficacy had a positive correlation with several domains of the AI-Technology

Acceptance Model (extended TAM), which include Personal Competence (extended TAM-PC), Social Influence

(extended TAM-SI), Perceived AI Trust (extended TAM-PAI), Perceived Usefulness of AI (extended TAM-AIU,

efficacy is a strong predictor of AI dependence, consistent with the relationships identified in this study. (Morales

García et al. 2024)

acceptance. The detailed correlation Tables 3.1 and 3.2 reveal that DAI was significantly correlated with

numerous other extended TAM constructs with comparable or stronger magnitudes. For example, extended

TAM-PC and extended TAM-PAI also showed significant positive correlations with DAI, emphasizing various

factors linked to AI-dependence. Consistent with prior research on technology acceptance, our study reveals a

positive correlation between greater AI reliance and users' positive attitudes toward its utility and acceptance

(Davis, 1989; Hoffman et al., 2018). Essentially, when individuals perceive AI as both beneficial and reliable,

they instinctively incorporate it more deeply into their workflows. However, this advantageous integration comes

with a notable warning. Although heightened trust is typically favorable, it may inadvertently diminish when the

technology’s functions are influenced by human biases, which could result in an unhealthy dependency that

undermines sound decision-making, as explored in relation to suitable levels of automation dependency that

emerged as a significant positive predictor of AI dependence (β = 0.242, p<0.01). This suggests a direct relation:

the more human-like people perceive AI to be, the more they tend to rely on it. This finding aligns well with the

common psychological tendency for people to connect with or lean more heavily on things that show human-

like traits. Various studies on Chatbots have found similar results in which users are more likely to rely on AI

chatbots when they exhibit human-like traits (Moriuchi, E.,2021; Cheng, S., et al., 2022 ).

Conversely, Perceived AI Trust (extended TAM-PAI) was found to be a significant negative predictor of AI

dependence (β = -0.433, p = .002). This outcome might seem unexpected: it suggests that when people have

more trust in AI's ability to work well and reliably, they depend on it less. One way to understand this is that

individuals who truly trust AI see it as a dependable tool that helps them achieve more, rather than something

they need to constantly check or lean on excessively. Studies on human automation interaction have found that

detrimental over-dependency that could undermine an individual's autonomy and self-confidence.

It is important to note that general self-efficacy (NGSES) and various subcomponents of AI self-efficacy, such

as Assistance (AISE-AS) and Comfort with AI (AISE-CF), while showing a strong correlation with DAI in

bivariate evaluations, did not appear as significant independent factors in the multivariate regression analysis.

This indicates that their impact on DAI may be influenced by other factors incorporated in the broader model.

Attachment Styles and AI Dependence A crucial finding was the absence of significant correlations between

both Avoidance (ECR-AO) and Anxiety (ECR-AX) attachment styles and AI dependence (DAI), as well as other

AI-related self-efficacy or extended TAM constructs. Moreover, neither attachment style significantly predicted

DAI in the regression analysis.This suggests that, at least within this specific group of Indian college students

and using the direct assessments applied, prevalent adult attachment styles do not seem to have a straightforward

This conclusion, while seemingly contrary to some intuitive expectations, resonates with the intricate insights of

attachment theory in new scenarios. Traditional attachment theory mainly focuses on human-to-human

relationships (Bowlby, 1969; Fraley et al., 2000), yet recent investigations have begun to assess its relevance in

In contrast, avoidant attachment has occasionally shown little to no significant impact on the adoption of or trust

in AI (Gillath, 2021; Wu et al., 2025).

It is essential to distinguish between adult attachment style and the development of particular "attachment-like"

tendencies towards AI. While the measures for general adult attachment in our study (ECR-AO, ECR-AX) did

not directly determine AI dependence, this does not rule out the potential for AI to serve specific attachment-

related roles for some individuals, as proposed by new scales intended to evaluate experiences in human-AI

relationships (Wu et al,2025). The lack of a direct linear correlation in our model may suggest that the effects of

attachment styles on AI dependence are more intricate, possibly mediated by alternate psychological factors or

specific interaction behaviors not captured during the assessment.

systems. The finding that human-like features in AI predict greater dependence suggests a key challenge: while

these features can certainly make AI more engaging, they might also, by accident, lead to users relying on it too

much. Therefore, those who create AI should carefully think about the right balance. They need to weigh making

AI easy and pleasant to use through these human-like traits against the risk of encouraging an unhealthy level of

reliance.

On the other hand, actively building trust in AI systems and clearly communicating that trust could be a good

way to reduce problematic dependence. This would help users see AI as a dependable tool that gives them power,

rather than something they feel forced to use all the time. Teachers and policymakers could use these ideas to

create specific programs. These programs would aim to help students use AI in a healthy and balanced way, build

critical understanding of AI, and make sure that using AI actually improves, rather than reduces, their

chance of common method bias, where the way the data was collected might influence the results. Future studies

could get around this by using more objective ways to measure AI use or by observing people's behavior, not

just asking questions.

Third, the sample only included college students in India. This means findings might not apply to people in other

age groups, different cultures, or various job settings. Future research should aim to include more varied and

representative groups of people.